The Haunted DAG & The Causal Terror
library(here)
source(here::here("code/scripts/source.R"))
slides_dir = here::here("docs/slides/L06")
slides_dir = here::here("docs/slides/L07")




The classic confound involves some exposure $E$. A lot of research goes into understanding what the returns on education are. There are a lot of confounds $U$. How do we de-confound this DAG using the back-door criterion? $E \leftarrow U \rightarrow W$. How to shut this? It's a fork, and you close it by conditioning on $U$. But here it's unobserved, so we can't condition on it. You therefore can't get an unbiased estimate. But that's an achievement, because we've wasted less time.
don't know how to handle 'marginfigure' engine output
